JICDA: Journal of Informatics, Computer Science, Data Science And Artificial Intelligence

Dental Disease Classification Using Image Data And Machine Learning Models

Agus Fahmi Limas Ptr — 2025-06-20

This study investigates the application of machine learning techniques for the classification of dental diseases based on image data. Two models—Naive Bayes and Neural Network—were evaluated using a publicly available dataset containing 13,839 annotated images across ten dental disease categories. Image embeddings were extracted using the pre-trained Inception v3 model to convert raw images into structured feature vectors. These features were then used to train and evaluate both classifiers using standard performance metrics, including AUC, precision, recall, and F1-score. The results indicate a significant performance gap between the two models. The Neural Network outperformed Naive Bayes across all metrics, achieving an AUC of 0.932 and an F1-score of 0.669, while Naive Bayes performed poorly with near-zero precision and recall. Confusion matrix analysis further highlighted the Neural Network’s superior ability to handle multiclass classification, although it still struggled with underrepresented classes such as Caries 2, Caries 3, and Caries 4. These findings suggest that deep learning-based approaches, when combined with robust image embeddings, are more effective for dental disease classification tasks and offer strong potential for supporting automated diagnostic systems in dentistry.

RSA ENCRYPTION FOR DATA SECURITY IN QR CODE BASED DIGITAL PAYMENT SYSTEM

Muhammad Khalid Hakim Manulang — 2025-06-22

This research, titled "RSA Encryption for Securing Data in QR Code-Based Digital Payment Systems," aims to enhance transaction data security in the increasingly popular digital payment environment. In today’s digital era, the use of cashless payment methods through applications such as OVO, DANA, and GoPay simplifies transactions but also introduces risks related to data security. Therefore, a solution is needed to ensure the confidentiality and integrity of data. In this study, the RSA (Rivest-Shamir-Adleman) algorithm is employed to encrypt transaction data before it is embedded in a QR Code format. This process involves the generation of public and private keys, data encryption, and the creation of a unique QR Code for each transaction. The results indicate that using RSA with a 2048-bit key length provides a high level of security, effectively protecting data from unauthorized access and ensuring data integrity throughout the transmission process.Based on the analysis, the integration of RSA encryption and QR Code technology proves to be effective in mitigating data theft risks. This research is expected to serve as a reference for the further development of more secure digital payment applications, while also offering insights into the real-world application of cryptography.

Keywords : Digital Payment, QR Code, RSA, Key Generator

IMPELEMENTATION OF DSA ALGORITHM IN DIGITAL DOCUMENT SECURITY

Annisah Amalia — 2025-06-07

The Digital Signature Algorithm (DSA) is a cryptographic method used to ensure the integrity and authenticity of data by generating a unique digital signature for each document. In this study, the author examines the implementation steps of DSA, including the generation of public and private keys, as well as how digital signatures can be used to verify the sender's identity and prevent forgery. The implementation results indicate that the use of DSA significantly enhances the security of digital documents, providing strong protection against data tampering. These findings are expected to contribute to the development of more effective and reliable information security systems in the digital era.

Keywords : Cryptography, Hash Function, Asymmetric Key, Digital Signature Algorithm, Secure Hash Algorithm-256

CNN-Based Pet Image Classification: A Deep Learning Approach with Orange

Irwan Daniel — 2025-08-06

Image classification is a key task in computer vision, with Convolutional Neural Networks (CNNs) excelling at recognizing complex patterns. This study focuses on the binary classification of cat and dog images using Orange Data Mining, a visual programming platform that enables machine learning without coding. A balanced dataset of 1,000 labeled images (500 cats, 500 dogs) from Kaggle was used. Images were embedded into feature vectors using a pretrained ResNet50 model, then classified using two models: a Neural Network (CNN-based) and Naive Bayes. The performance was evaluated using accuracy, precision, recall, F1-score, and a confusion matrix. The CNN model achieved 98% accuracy, with 99.18% precision, 96.80% recall, and an F1-score of approximately 97.9%. Only 20 images were misclassified, indicating strong generalization and low bias. These results confirm the effectiveness of CNNs for pet image classification and demonstrate the value of using pretrained embeddings like ResNet50. The study also highlights Orange’s suitability for deep learning by offering an accessible, low-code environment. This approach is ideal for educational use, prototyping, or real-world applications such as pet recognition and smart surveillance. The findings support broader use of visual programming tools in democratizing AI development.

Keywords: Image Classification, CNN, Orange, Dogs, Cats, Machine Learning

Classification Of Oil Palm Fruit Ripeness Level Using Artificial Neural Network

Aulia Ichsan Aulia Ichsan — 2025-06-30

The manual sorting process for determining the ripeness of oil palm fruit is subjective and inefficient, leading to a decline in Crude Palm Oil (CPO) quality and economic losses. This study aims to develop an automatic classification system for oil palm fruit ripeness to address these issues. It employs a digital image processing approach using a Neural Network model. The methodology involves using a pre-trained InceptionV3 model for feature extraction from a dataset of 3,000 fruit images, which are then fed into a custom-designed neural network with three hidden layers, using ReLU as the activation function and Adam as the optimizer. The model successfully classifies the fruits into 'unripe', 'ripe', and 'overripe' categories. The results show a high overall accuracy of 96.56 percent, with an F1-Score of 96.55 percent. The study concludes that the proposed Neural Network model is highly effective and reliable for automating oil palm fruit sorting, offering a feasible solution to improve efficiency and standardization in the palm oil industry

Keywords: Oil Palm Fruit, Ripeness Classification, Neural Network, Image Processing, InceptionV3

IMPLEMENTATION OF DEEP LEARNING ALGORITHM IN HANDWRITING TO TEXT DOCUMENT CONVERSION APPLICATION

Alif Lufti — 2025-06-27

The development of information and communication technology has driven the need for systems capable of efficiently converting handwritten text into digital text. This study aims to develop a web-based application capable of real-time handwriting recognition using Tesseract.js, a JavaScript library for optical character recognition (OCR). The application is designed to assist users in converting handwritten documents into editable text formats, thereby enhancing productivity and information accessibility.

The methods used in this study include uploading handwritten images, preprocessing the images to improve input quality, and applying OCR algorithms using Tesseract.js to recognize characters and words. The recognized results are then displayed on the user interface, with an option for manual correction if needed. The study also evaluates the accuracy of the text recognition produced by the application by comparing the recognition results with the original text.

The results show that the developed application is capable of recognizing handwriting with a satisfactory level of accuracy, despite variations in handwriting styles. This application is expected to make a significant contribution in the field of document digitization and data processing, and serve as a reference for the development of similar systems in the future.

Padang Cuisine Classification Using Resnet Features And Transfer Learning

Victor Ginting — 2025-06-23

Padang cuisine is a renowned Indonesian culinary heritage with diverse dishes that are often challenging to identify visually due to their similarities and variations. This study proposes a digital image-based classification system to accurately recognize five types of Padang dishes: Rendang, Dendeng Batokok, Ayam Pop, Gulai Tambusu, and Gulai Tunjang. Feature extraction was performed using a pre-trained ResNet model to leverage its deep residual architecture, which effectively captures rich visual information. The extracted features were then used to train and compare two classifiers: Support Vector Machine (SVM) with Radial Basis Function (RBF) kernel, and Multi-Layer Perceptron (MLP) with three hidden layers and ReLU activation optimized by Adam. The dataset comprised 548 images collected from Kaggle, evenly distributed across the five dish categories. Evaluation metrics included accuracy, precision, and recall. Experimental results show that SVM achieved the highest performance with an accuracy of 87.9%, outperforming MLP, which obtained 87.3%. The findings suggest that SVM with RBF kernel is more suitable for classifying Padang dishes on a limited dataset, while MLP holds potential with further optimization and larger datasets. This research contributes to the advancement of automated culinary recognition systems, supporting the preservation and promotion of Indonesian culinary heritage through artificial intelligence.

Keywords : Padang Cuisine Classification, Transfer Learning, ResNet, Support Vector Machine (SVM), Multi-Layer Perceptron (MLP)